Sustainable Differentially Reliable Architecture for Dark Silicon

1. A method comprising: calculating, by use of a processor, an expected energy efficiency for a prior mapping of process threads for a plurality of cores; calculating current reliability levels for the plurality of cores from degradation rates for the plurality of cores; calculating desired reliability levels for the plurality of cores from the current reliability levels and initial reliability levels for the plurality of cores; calculating an error desired level from the current reliability levels and the desired reliability levels; calculating a workload acceptance capacity (WAC) from the error desired level; and mapping the process threads to the plurality of cores, wherein a process thread is mapped to a core if a mapping policy is satisfied by a utilization for the core not exceeding an energy delay and the WAC for the core not reaching a specified WAC threshold and a specified number of the plurality of cores are not powered and are not mapped to any of the process threads.

2. The method of claim 1 , wherein the process thread is mapped to the core if the mapping policy is further satisfied when the expected energy efficiency is optimal.

3. The method of claim 1 , wherein the process thread is mapped to the core if the mapping policy is further satisfied when the WAC for the core exceeds the WAC threshold.

4. The method of claim 1 , wherein the process thread is mapped to the core if the mapping policy is further satisfied in response to the expected energy efficiency exceeding an efficiency threshold for the core, wherein the WAC for the core exceeds the WAC threshold.

5. The method of claim 1 , wherein the WAC w(t) is calculated as w(t)=(1−γ)1−K p ed(t), where ed(t) is the error desired level between a desired reliability level d(t) and a current reliability level r(t), t is the process thread, and K p is a proportional gain constant.

6. The method of claim 1 , wherein the desired reliability level d(t) is calculated as d ⁡ ( t ) = r ⁡ ( 0 ) - ∑ i = 1 n ⁢ ⁢ er i ⁡ ( t ) ⁢ x ⁢ 1 - r ⁡ ( 0 ) n - ∑ i = 1 n ⁢ r i ⁡ ( 0 ) where r( 0 ) is an initial reliability level, er(t) is a reliability error vector between a current reliability level r(t) and the initial reliability level r( 0 ), 1 is a ones vector, and t is the process thread.

7. The method of claim 6 , wherein the mapping policy is further satisfied if a utilization for the core does not exceed an energy delay for the core.

8. The method of claim 7 , wherein the process threads are not mapped to the core if the core has reached the WAC threshold.

9. The method of claim 6 , wherein the current reliability level r(t) is a probability of a timing violation based on a process variation and an aging model.

10. The method of claim 1 , wherein the process thread is mapped to the core if an inequality ed(t) 2 >AW×(util(l)−WAC(l)) is satisfied, where ed(t) is an error desired vector, AW is a specified aging weight, util(l) is a utilization of the core, and WAC(l) is the WAC for the core.

11. The method of claim 1 , wherein the energy delay edl is calculated as edl=e*d 2 where e is an estimated energy for the core and d is a delay for the core.

12. An apparatus comprising: a non-transitory memory storing computer readable code executable by a processor, the computer readable code comprising: computer readable code that calculates an expected energy efficiency for a prior mapping of process threads for a plurality of cores, calculates current reliability levels for the plurality of cores from degradation rates for the plurality of cores, calculates desired reliability levels for the plurality of cores from the current reliability levels and initial reliability levels for the plurality of cores, calculating an error desired level from the current reliability levels and the desired reliability levels and calculating a workload acceptance capacity (WAC) from the error desired level; and computer readable code that maps the process threads to the plurality of cores, wherein a process thread is mapped to a core if a mapping policy is satisfied by a utilization for the core not exceeding an energy delay and the WAC for the core not reaching a specified WAC threshold and a specified number of the plurality of cores are not powered and are not mapped to any of the process threads.

13. The apparatus of claim 12 , wherein the process thread is mapped to the core if the mapping policy is further satisfied when the expected energy efficiency is optimal.

14. The apparatus of claim 12 , wherein the process thread is mapped to the core if the mapping policy is further satisfied when the WAC for the core exceeds the specified WAC threshold.

15. The apparatus of claim 12 , wherein the process thread is mapped to the core if the mapping policy is further satisfied in response to the expected energy efficiency exceeding an efficiency threshold for the core, wherein the WAC for the core exceeds the WAC threshold.

16. The apparatus of claim 12 , wherein the energy delay edl is calculated as edl=e*d 2 where e is an estimated energy for the core and d is a delay for the core.

17. A program product comprising a non-transitory computer readable storage medium storing computer readable code executable by a processor to perform: calculating an expected energy efficiency for a prior mapping of process threads for a plurality of cores; calculating current reliability levels for the plurality of cores from degradation rates for the plurality of cores; calculating desired reliability levels for the plurality of cores from the current reliability levels and initial reliability levels for the plurality of cores; calculating an error desired level from the current reliability levels and the desired reliability levels; calculating a workload acceptance capacity (WAC) from the error desired level; and mapping the process threads to the plurality of cores, wherein a process thread is mapped to a core if a mapping policy is satisfied by a utilization for the core not exceeding an energy delay and the WAC for the core not reaching a specified WAC threshold and a specified number of the plurality of cores are not powered and are not mapped to any of process threads.

18. The program product of claim 17 , wherein the process thread is mapped to the core if the mapping policy is further satisfied when the expected energy efficiency is optimal.

19. The program product of claim 17 , wherein the process thread is mapped to the core if the mapping policy is further satisfied when the WAC exceeds for the core exceeds the WAC threshold.

20. The program product of claim 17 , wherein the process thread is mapped to the core if the mapping policy is further satisfied in response to the expected energy efficiency exceeding an efficiency threshold for the core, wherein the WAC for the core exceeds the WAC threshold.